Diagnostic accuracy of noninvasive steatosis biomarkers with magnetic resonance imaging proton density fat fraction as gold standard

Abstract

BACKGROUND

Nonalcoholic fatty liver disease (NAFLD) is the most common chronic liver disease. The accuracy of noninvasive biomarkers for detecting hepatic steatosis is still limited.

AIM

To assess the diagnostic performance of noninvasive steatosis biomarkers in diagnosing NAFLD using magnetic resonance imaging proton density fat fraction (MRI-PDFF) as the gold standard.

METHODS

A total of 131 suspected NAFLD patients (60% male, median age 36 years) undergoing MRI-PDFF were consecutively recruited from a tertiary hospital. Steatosis grades determined by MRI-PDFF were classified as none (< 5%), mild (5%-11%), moderate (11%-17%), and severe (≥ 17%). Six steatosis biomarkers were calculated according to clinical parameters and laboratory tests, including fatty liver index, hepatic steatosis index, ZJU index, Framingham steatosis index, triglycerides and glucose index, and visceral adiposity index. The accuracy of these biomarkers in detecting hepatic steatosis was evaluated using the area under the receiver operating characteristic curves (AUCs). The Youden index was used to determine the optimal cut-off for each biomarker. The linear trend analysis of each biomarker across the steatosis grades was conducted by Mantel-Haenszel χ² test. Spearman's rank correlation assessed the relationship between steatosis biomarkers and MRI-PDFF.

RESULTS

Steatosis grades based on MRI-PDFF prevalence were: None 27%, mild 40%, moderate 15% and severe 18%. Six steatosis biomarkers showed a linear trend across the steatosis grades and a significant positive correlation with MRI-PDFF. The six steatosis biomarkers demonstrated AUCs near 0.90 (range: 0.857-0.912, all P < 0.001) for diagnosing NAFLD by MRI-PDFF ≥ 5%. The optimal cut-offs showed sensitivity between 84.4%-91.7% and specificity between 71.4%-85.7%. The diagnostic performance of these biomarkers in detecting moderate-to-severe and severe steatosis was relatively weaker.

CONCLUSION

These noninvasive steatosis biomarkers accurately diagnosed NAFLD and correlated well with MRI-PDFF for detecting NAFLD, but they did not effectively detect moderate or severe steatosis.

Key Words: Nonalcoholic fatty liver disease; Diagnosis; Noninvasive biomarker; Magnetic resonance imaging proton density fat fraction; Chinese population

Core Tip: This is the first study to investigate the accuracy of noninvasive biomarkers for detecting nonalcoholic fatty liver disease (NAFLD) based on magnetic resonance imaging proton density fat fraction (MRI-PDFF) in the Chinese population, which is currently considered the gold standard in diagnosing hepatic steatosis. Two types of cut-offs (single optimal cut-off and dual cut-off) of six steatosis biomarkers were analyzed. These noninvasive steatosis biomarkers demonstrated great accuracy in diagnosing NAFLD (MRI-PDFF ≥ 5%) and exhibited a strong correlation with MRI-PDFF. However, they may be ineffective in detecting moderate or severe steatosis.

INTRODUCTION

Nonalcoholic fatty liver disease (NAFLD) is the leading cause of chronic liver conditions globally, impacting approximately 25% of adults. It is characterized by at least 5% hepatic steatosis without other causes of liver disease or significant alcohol consumption[1,2]. NAFLD associated with obesity and insulin resistance (IR) as a liver manifestation of metabolic syndrome can progress to NASH, cirrhosis, hepatocellular carcinoma, and liver-related death, while also elevating the risk and mortality of cardiovascular events[2-4].

Liver biopsy is the clinical gold standard for diagnosing and grading hepatic steatosis. However, owing to its invasiveness, high cost, sample errors, and observer-dependence, it is not practical for the screening or monitoring of a population[5]. Recently, magnetic resonance imaging (MRI) techniques such as magnetic resonance spectroscopy (MRS) and MRI-proton density fat fraction (MRI-PDFF) have become precise and dependable noninvasive methods for measuring hepatic steatosis in NAFLD[6]. Nevertheless, MRS primarily serves as a research tool due to its limited clinical accessibility and the need for specialized expertise[7]. MRI-PDFF significantly correlates with MRS[8] and histology-confirmed steatosis grade[9-11], defined as the ratio of mobile proton density from triglycerides to the total proton density from mobile triglycerides and water[12]. MRI-PDFF ≥ 5% has been extensively used as a diagnostic or inclusion criterion for NAFLD in numerous studies and trials[13-16]. Moreover, Noureddin et al[17] demonstrated that MRI-PDFF is more effective than histology in detecting minor variations in liver fat content. However, MRI-PDFF remains expensive and not routinely accessible. Consequently, there is growing interest in creating noninvasive, straightforward, precise, cost-effective, and accessible techniques for early steatosis detection, quantification, and post-treatment monitoring[18].

Several noncommercial biomarkers derived from routine clinical data, including the fatty liver index (FLI), hepatic steatosis index (HSI), ZJU index, Framingham steatosis index (FSI), triglycerides and glucose (TyG) index, and visceral adiposity index (VAI), have demonstrated strong efficacy in detecting hepatic steatosis[19-24]. However, the key drawbacks of these noninvasive steatosis biomarkers were the utilization of a suboptimal reference standard by ultrasonography rather than liver biopsy or MRI-PDFF, and lack of independent external validation[6,18]. A French retrospective study of 324 liver biopsies, where only 5% lacked hepatic steatosis, confirmed the diagnostic accuracy of FLI, HSI, and VAI indices for detecting steatosis[18]. However, their performance relative to MRI-PDFF as the reference standard in the Chinese population remains uncertain.

The aim of this study was to assess and compare the diagnostic performance of the six noninvasive steatosis biomarkers (FLI, HSI, ZJU index, FSI, TyG index, and VAI) in identifying NAFLD as determined by MRI-PDFF in a Chinese adult population, including their ability to predict moderate or severe steatosis according to the MRI-PDFF results. Two types of cut-offs were identified. The first one was a single optimal cut-off to distinguish subjects with and without NAFLD. Then, owing to the dual cut-offs proposed by many previous studies[19,20,22], we also identified the dual cut-offs that have the best accuracy to either exclude or confirm NAFLD.

MATERIALS AND METHODS

Study design

A single-center observational study was conducted on adults suspected of having NAFLD who underwent MRI. Participants were consecutively enrolled during annual health check-ups at the China-Japan Friendship Hospital in Beijing, China. From January 2019 to April 2019, a total of 203 potential eligible subjects were screened, 186 of which were deemed eligible for this study. Finally, 131 subjects underwent MRI-PDFF within a 7-day period after entry (Supplementary Figure 1). This study protocol adhered to the ethical principles of Declaration of Helsinki and received approval from the Ethics Committee of China-Japan Friendship Hospital in Beijing, China. All participants provided written informed consent prior to inclusion.

Inclusion and exclusion criteria

Inclusion criteria[9] were as follows: age > 18 years and ≤ 65 years, willingness and ability to finish all procedures in the protocol, and signed the informed consent. Exclusion criteria[13] were as follows: Alcohol consumption ≥ 210 g/week in men or ≥ 140 g/week in women within 10 years of recruitment; presence of secondary hepatic steatosis causes, such as total parenteral nutrition or medications inducing secondary steatosis; and evidence of other liver diseases, including viral hepatitis, inherited metabolic liver disorders, autoimmune liver diseases, drug-induced liver injury, and heart failure. Participants were excluded if they had cirrhosis, liver cancer, decompensated liver disease, dysfunction of vital organs, significant systemic diseases, were pregnant or attempting pregnancy, had contraindications to MRI, or lacked essential data.

Data collection

All subjects were invited for a face-to-face visit. Age, gender, smoking status, alcohol consumption, history of diabetes, hypertension, other severe diseases, and related medications were obtained with a standard questionnaire. Body mass index (BMI) was calculated by dividing body weight by the square of standing height. Patients were categorized based on the Chinese Working Group on Obesity criteria: Underweight (< 18.5 kg/m²), normal weight (18.5-24 kg/m²), overweight (24-28 kg/m²), and obese (≥ 28 kg/m²)[25]. Waist circumference was assessed using a non-flexible tape at the midpoint between the rib cage's lower border and the iliac crest. After a 5-minute rest in a seated position, blood pressure was consecutively measured three times at 1-minute intervals using an OMRON HBP-9021 automated electronic device (Omron, Japan), and the average of these readings was used for analysis. Metabolic syndrome was identified based on the Chinese Diabetes Society criteria[26], requiring at least three of the following conditions: abdominal obesity (waist circumference ≥ 90 cm for males, ≥ 85 cm for females), arterial hypertension (systolic blood pressure ≥ 130 mmHg, diastolic blood pressure ≥ 80 mmHg, or use of anti-hypertensive medication), hyperglycemia (fasting blood glucose ≥ 6.1 mmol/L or previously diagnosed type 2 diabetes mellitus under treatment), elevated fasting serum triglycerides (≥ 1.70 mmol/L), and reduced fasting serum high-density lipoprotein cholesterol (< 1.04 mmol/L). A fasting blood sample was collected on the day of inclusion after a minimum of 8 hours fasting overnight to assess biochemical marker levels.

MRI protocol

To measure liver PDFF, an advanced MRI technique was used[9,17]. Please see Supplementary Material for further details.

Definition of NAFLD

NAFLD was defined by an MRI-PDFF ≥ 5% in the absence of other hepatic steatosis causes[13-16], while non-NAFLD control was defined by an MRI-PDFF of < 5%. To determine the diagnostic accuracy of noninvasive steatosis biomarkers for predicting moderate or severe steatosis, we also set ≥ 11% and ≥ 17% as reference thresholds according to the previous study by Imajo et al[10]. Hepatic steatosis, as evaluated by MRI-PDFF, was categorized into four ordinal grades: None (< 5%), mild (5%-11%), moderate (11%-17%), and severe (≥ 17%). PDFF maps examples are shown in Figure 1.

Figure 1 Examples of magnetic resonance imaging fat fraction maps for steatosis grade by proton density fat fraction. A: A 32-year-old woman with no steatosis (nonalcoholic fatty liver disease); B: A 34-year-old woman with mild steatosis; C: A 29-year-old woman with moderate steatosis; D: A 33-year-old man with severe steatosis. The average of the proton density fat fraction calculated by the region-of interests of each liver segment is at the bottom of each figure.

Noninvasive steatosis biomarkers

According to previously published formulas[19-24] (Table 1), six liver steatosis biomarkers (the FLI, HSI, ZJU index, FSI, TyG index, and VAI) were measured. Two validated noninvasive serum biomarkers, the NAFLD fibrosis score (NFS) and fibrosis-4 index (FIB-4), were calculated to estimate liver fibrosis severity without a biopsy, as previously reported[27] (Table 1). The exclusionary and confirmatory cut-off value of NFS and FIB-4 for predicting advanced fibrosis in NAFLD patients was < -1.455 and > 0.676, < 1.30 and > 2.67, respectively.

Table 1 Noninvasive biomarkers for predicting hepatic steatosis and fibrosis in this study.

Biomarkers	Component and formula
Steatosis biomarkers
FLI	FLI = 100/ (1 + e-z); Z = 0.953 × lnTG (mg/dL) + 0.139 × BMI (kg/m2) + 0.718 × lnGGT (IU/L) + 0.053 × WC (cm) - 15.745
HSI	HSI = 8 × ALT/AST + BMI (kg/m2) + 2 (if diabetes) + 2 (if female)
ZJU index	ZJU index = BMI (kg/m2) + FBG (mmol/L) + TG (mmol/L) + 3 × ALT/AST + 2 (if female)
FSI	FSI = 100/ (1 + e-z); Z = -7.981 + 0.011 × age (years) – 0.146 × sex (female = 1, male = 0) + 0.173 × BMI + 0.007 × TG (mg/dL) + 0.593 × hypertension (yes = 1, no = 0) + 0.789 × diabetes (yes = 1, no = 0) + 1.1 × ALT/AST ratio ≥ 1.33 (yes = 1, no = 0)
TyG index	TyG index = ln (TG (mg/dL) × FBG (mg/dL)/2)
VAI	VAI (male)= [WC (cm)/39.68 + (1.88 × BMI (kg/m2))] × (TG (mmol/L)/1.03) × (1. 31/HDL (mmol/L)); VAI (female) = [WC (cm)/36.58 + (1.89 × BMI (kg/m2))] × (TG (mmol/L)/0.81) × (1.52/HDL (mmol/L))
Fibrosis biomarkers
NFS	NFS = -1.675 + 0.037 × age (years) + 0.094 × BMI (kg/m2) + 1.13 × (hyperglycemia; yes = 1, no = 0) + 0.99 × AST/ALT ratio - 0.013 × platelet (× 109/L) - 0.66 × albumin (g/dL)
FIB-4	FIB-4 = age (years)× AST (IU/L)/platelet (× 109/L) × ALT (IU/L)1/2

FLI: Fatty liver index; HSI: Hepatic steatosis index; FSI: Framingham steatosis index; TyG: Triglycerides and glucose; VAI: Visceral adiposity index; NFS: Nonalcoholic fatty liver disease fibrosis score; FIB-4: Fibrosis-4 index; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase; BMI: Body mass index; TG: Triglyceride; GGT: Gamma-glutamyl transpeptidase; HDL: High-density lipoprotein; WC: Waist circumference.

Statistical analysis

Statistical analyses were conducted using IBM SPSS version 20.0 (IBM Corp., Armonk, NY). A two-tailed P < 0.05 was considered statistically significant. Continuous variables were presented as mean ± SD or median with IQR. Categorical variables were presented as frequencies and percentages. To compare baseline continuous variables across ordinal steatosis grades based on MRI-PDFF results, a one-way analysis of variance (ANOVA) with trend analysis[18] was used for normally distributed data with equal variance; otherwise, the Jonchheere-Terpstra trend test was applied[28]. The Mantel-Haenszel χ² test for linear trend analysis[29] was used for categorical variables among the different steatosis grades. Spearman's rank correlation coefficient (r) was used to assess the relationship between steatosis biomarkers and MRI-PDFF. The receiver operating characteristics (ROCs) curves were performed to assess diagnostic accuracy of the noninvasive steatosis biomarkers. The areas under the ROC curves (AUCs) with 95% confidence interval (CI) were recorded.

Single optimal cut-off values for NAFLD were identified by the maximum Youden index analysis. Dual cut-off values were determined using established criteria[30,31] to optimally confirm or exclude NAFLD (MRI-PDFF ≥ 5%). We established the confirmatory cut-off with a specificity greater than 90% and a positive likelihood ratio (PLR) of at least 10, as well as the exclusionary cut-off with a sensitivity exceeding 90% and a negative likelihood ratio (NLR) of 0.1 or less[30,31]. For each cut-off value, the AUC, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), PLR, NLR, and diagnostic accuracy were calculated.

For sample size estimation, based on the diagnostic accuracy of HSI with liver biopsy as the reference standard[18], the AUC for hepatic steatosis was anticipated to be at least 0.80, with values below 0.6 considered the null hypothesis. In addition, a dropout rate of less than 10% was predicted. To achieve a power of 90% with an α of 0.05 and a 1:3 ratio of the sample sizes in the non-NAFLD groups to the NAFLD groups, a projected sample size of at least 92 participants was needed. Thus, the study total of 35 non-NAFLD subjects and 96 NAFLD subjects was sufficiently powered to test whether the AUC of HSI for hepatic steatosis was higher than 0.80.

RESULTS

Baseline characteristics

In this observational study, the median age was 36 (IQR: 32-47) years (range: 23-65 years), and 78 (59.5%) of the 131 participants undergoing MRI-PDFF were male. Table 2 presents the baseline characteristics categorized by steatosis grade as determined by MRI-PDFF. All MRI-PDFFs were successfully performed without technical issues. The median time between MRI and the blood sample draw or clinical interview was 1 day. In terms of MRI-PDFF, 26.7% (n = 35) of patients showed no steatosis (MRI-PDFF < 5%), while mild, moderate, and severe steatosis were observed in 36.6% (n = 48), 19.1% (n = 25), and 17.6% (n = 23) of patients, respectively.

Table 2 Baseline characteristics.

Characteristics	Total	None (PDFF < 5%)	Mild (5% ≤ PDFF < 11%)	Moderate (11% ≤ PDFF < 17%)	Severe (PDFF ≥ 17%)	P value for trend3
Number	131 (100)	35 (26.7)	48 (36.6)	25 (19.1)	23 (17.6)
MRI-PDFF	9.6 ± 7.4	2.5 ± 0.9	7.1 ± 1.8	12.5 ± 1.7	22.7 ± 5.1	< 0.001
Median (IQR)	7.7 (4.1-12.1)	2.3 (1.8-3.3)	6.9 (5.5-8.7)	11.9 (11.4-13.9)	20.9 (18.9-25.5)	< 0.001
Demographics
Age, years	36 (32-47)	34 (31-38)	39 (33-51)	42 (35-54)	33 (29-42)	0.487
Male	78 (59.5)	13 (37.1)	29 (60.4)	16 (64.0)	20 (87.0)	< 0.001
Current smoking	25 (19.1)	5 (14.3)	11 (22.9)	6 (24.0)	3 (13.0)	0.978
Anthropometrics
BMI, kg/m2	27 ± 4	24 ± 3	27 ± 3	28 ± 3	30 ± 3	< 0.001
Overweight1	105 (80.2)	17 (48.6)	42 (87.5)	23 (92.0)	23 (100.0)	< 0.001
Obesity2	43 (32.8)	5 (14.3)	10 (20.8)	12 (48.0)	16 (69.6)	< 0.001
Waist circumference, cm	94.0 ± 10.8	85.9 ± 11.2	95.0 ± 8.8	99.8 ± 8.8	101.8 ± 6.8	< 0.001
Waist-to-hip ratio	0.90 ± 0.06	0.86 ± 0.06	0.91 ± 0.04	0.94 ± 0.05	0.94 ± 0.04	< 0.001
Metabolic diseases
Abdominal obesity	99 (75.6)	15 (42.9)	37 (77.1)	25 (100.0)	22 (95.7)	< 0.001
Arterial hypertension	68 (51.9)	12 (34.3)	25 (52.1)	17 (68.0)	14 (60.9)	0.017
Type 2 diabetes mellitus	12 (9.2)	2 (5.7)	3 (6.2)	5 (20.0)	2 (8.7)	0.283
Hyperglycemia	19 (14.5)	2 (5.7)	6 (12.5)	8 (32.0)	3 (13.0)	0.108
High TG	69 (52.7)	6 (17.1)	29 (60.4)	18 (72.0)	16 (69.6)	< 0.001
Low HDL-C	33 (25.2)	6 (17.1)	15 (31.2)	6 (24.0)	6 (26.1)	0.573
Metabolic syndrome	58 (44.3)	8 (22.9)	21 (43.8)	18 (72.0)	11 (47.8)	0.007
Laboratory tests
ALT, IU/L	28 (19-43)	14 (11-23)	26 (21-38)	34 (27-49)	53 (37-100)	< 0.001
AST, IU/L	22 (18-26)	18 (16-22)	21 (17-26)	22 (20-26)	31 (24-40)	< 0.001
Total bilirubin, μmol/L	13.5 ± 6.4	12.5 ± 6.0	13.0 ± 5.0	14.9 ± 6.2	14.2 ± 8.9	0.292
GGT, IU/L	29 (21-50)	17 (11-25)	29 (22-50)	28 (23-55)	50 (35-72)	< 0.001
Albumin, g/L	47 ± 2	47 ± 3	48 ± 2	47 ± 2	48 ± 2	0.519
TC, mmol/L	4.61 ± 0.86	4.26 ± 0.77	4.76 ± 0.86	4.71 ± 0.96	4.71 ± 0.81	0.073
TG, mmol/L	1.87 ± 1.20	0.96 ± 0.56	2.02 ± 1.04	2.09 ± 1.05	2.70 ± 1.54	< 0.001
HDL-C, mmol/L	1.24 ± 0.29	1.38 ± 0.37	1.22 ± 0.27	1.19 ± 0.22	1.12 ± 0.17	0.001
LDL-C, mmol/L	2.76 ± 0.77	2.45 ± 0.59	2.92 ± 0.78	2.87 ± 0.91	2.80 ± 0.73	0.126
FBG, mmol/L	5.23 (4.93-5.81)	4.94 (4.68-5.31)	5.23 (5.02-5.81)	5.50 (4.97-6.16)	5.44 (5.21-5.93)	< 0.001
Uric acid, μmol/L	343 ± 88	299 ± 99	344 ± 66	353 ± 86	397 ± 86	< 0.001
Platelet count, × 109/L	265 ± 60	255 ± 64	265 ± 61	270 ± 52	273 ± 61	0.243
Steatosis biomarkers
FLI	53 ± 28	24 ± 22	56 ± 21	66 ± 17	78 ± 18	< 0.001
HSI	39 ± 7	33 ± 5	39 ± 5	42 ± 4	47 ± 6	< 0.001
ZJU index	39 ± 5	34 ± 4	40 ± 4	42 ± 3	45 ± 4	< 0.001
FSI	30 (14-55)	6 (3-15)	29 (20-42)	52 (37-64)	62 (50-87)	< 0.001
TyG index	8.8 ± 0.7	8.1 ± 0.5	9.0 ± 0.5	9.0 ± 0.5	9.3 ± 0.5	< 0.001
VAI	2.07 (1.33-3.17)	1.04 (0.71-1.44)	2.25 (1.66-3.25)	2.28 (1.76-3.88)	2.97 (2.02-4.88)	< 0.001
Fibrosis biomarkers
NFS	-3.204 ± 1.001	-3.231 ± 1.111	-3.205 ± 1.096	-2.950 ± 1.134	-3.593 ± 1.066	0.316
NFS < -1.455	125 (95.4)	33 (94.3)	47 (97.9)	23 (92.0)	22 (95.7)	0.888
-1.455 ≤ NFS ≤ 0.676	6 (4.6)	2 (5.7)	1 (2.1)	2 (8.0)	1 (4.3)
NFS > 0.676	0	0	0	0	0
FIB-4	0.58 (0.48-0.76)	0.69 (0.52-0.76)	0.65 (0.44-0.82)	0.56 (0.47-0.91)	0.49 (0.42-0.67)	0.239
FIB-4 < 1.30	122 (93.1)	34 (97.1)	42 (87.5)	24 (96.0)	22 (95.7)	0.876
1.30 ≤ FIB-4 ≤ 2.67	9 (6.9)	1 (2.9)	6 (12.5)	1 (4.0)	1 (4.3)
FIB-4 > 2.67	0	0	0	0	0

¹Overweight was defined as body mass index ≥ 24 kg/m².

²Obesity was defined as body mass index ≥ 28 kg/m².

³P value for trend was determined by comparing the baseline characteristics of subjects from four ordinal steatosis grade groups (none, mild, moderate, severe) using a one-way analysis of variance for trend analysis, Jonchheere-Terpstra trend test, or Mantel-Haenszel χ² test for linear trend analysis.

Data are expressed as the mean ± SD or median (IQR) or n (%). PDFF: Proton density fat fraction; MRI-PDFF: Magnetic resonance imaging proton density fat fraction; BMI: Body mass index; TG: Triglyceride; TC: Total cholesterol; HDL-C: High-density lipoprotein cholesterol; LDL-C: Low-density lipoprotein cholesterol; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase; GGT: Gamma-glutamyl transpeptidase; FBG: Fasting blood glucose; FLI: Fatty liver index; HSI: Hepatic steatosis index; FSI: Framingham steatosis index; TyG: Triglycerides and glucose; VAI: Visceral adiposity index; NFS: Nonalcoholic fatty liver disease fibrosis score; FIB-4: Fibrosis-4 index.

The mean or median values of all steatosis biomarkers were as follow: FLI 53 ± 28, HSI 39 ± 7, ZJU index 39 ± 5, TyG index 8.8 ± 0.7, FSI 30 (IQR: 14-55), and VAI 2.07 (IQR: 1.33-3.17). All steatosis biomarkers exhibited a significant linear increase across varying steatosis grades as measured by MRI-PDFF (all P for trend < 0.001; see Figure 2 and Table 2). Figure 3 illustrates a strong positive correlation between six steatosis biomarkers and MRI-PDFF as a continuous variable, with Spearman’s rank correlation coefficients as follows: FLI (r² = 0.472), HSI (r² = 0.533), ZJU index (r² = 0.527), FSI (r² = 0.563), TyG index (r² = 0.338), and VAI (r² = 0.274), all with P < 0.001.

Figure 2 Distribution of each steatosis biomarkers according to steatosis grades as determined by magnetic resonance imaging proton density fat fraction. The box represents the interquartile range. The black line across the box represents the median. The whiskers represent the maximum and minimum values, excluding outliers (dots). P for trend was determined by one-way analysis of variance for trend analysis or Jonchheere-Terpstra trend test. MRI-PDFF: Magnetic resonance imaging proton density fat fraction.

Figure 3 Scatterplots show a correlation between the steatosis biomarkers and magnetic resonance imaging proton density fat fraction as a continuous variable. Correlation was evaluated by the Spearman correlation coefficient (r). MRI-PDFF: Magnetic resonance imaging proton density fat fraction.

In addition, no patients rose above the confirmatory advanced fibrosis cut-off (> 0.676 for NFS and > 2.67 for FIB-4) in Table 2, which suggests that the majority of NAFLD subjects had minimal or no fibrosis.

Diagnostic accuracy and single optimal cut-offs of biomarkers for the detection of NAFLD as defined by MRI-PDFF ≥ 5%

Figure 4A and Table 3 demonstrate that all steatosis biomarkers accurately differentiated between NAFLD (MRI-PDFF ≥ 5%) and non-NAFLD (MRI-PDFF < 5%). The FSI, ZJU index, and TyG index yielded the best AUCs of 0.912 (95%CI: 0.849-0.954), 0.903 (0.843-0.962), and 0.902 (0.839-0.966), respectively. The FLI, HSI, and VAI yielded AUCs of 0.899 (0.832-0.966), 0.898 (0.839-0.955), and 0.857 (0.774-0.937), respectively. There were no statistically significant differences in the AUCs among these models (all P > 0.05). Table 3 displays the sensitivity, specificity, PLR, NLR, PPV and NPV for assessing MRI-PDFF ≥ 5% with the single optimal cut-off as identified by the Youden index for each steatosis biomarker. The PPVs for predicting NAFLD (MRI-PDFF ≥ 5%) were nearly 90% for the following thresholds: FLI ≥ 42, HSI ≥ 34, ZJU index ≥ 37, FSI ≥ 15, TyG index ≥ 8.5, and VAI ≥ 1.49. The optimal cut-offs demonstrated sensitivity ranging from 84.4% to 91.7% and specificity from 71.4% to 85.7%.

Figure 4 Area under the receiver operating characteristic curves. A: The steatosis biomarkers for detecting nonalcoholic fatty liver disease [magnetic resonance imaging proton density fat fraction (MRI-PDFF) ≥ 5%]; B: The steatosis biomarkers for detecting moderate-to-severe steatosis (MRI-PDFF ≥ 11%); C: The steatosis biomarkers for detecting severe steatosis (MRI-PDFF ≥ 17%). AUC: Area under the receiver operating characteristic curve.

Table 3 The diagnostic accuracy of each steatosis biomarkers in identifying nonalcoholic fatty liver disease as defined by magnetic resonance imaging proton density fat fraction ≥ 5%.

Biomarkers	AUC (95%CI)	Cut-off1	Youden index	SE, %	SP, %	PLR	NLR	PPV, %	NPV, %	Accuracy, %
FLI	0.899 (0.832-0.966)	≥ 42	0.701	84.4	85.7	5.91	0.18	94.2	66.7	84.0
HSI	0.898 (0.839-0.955)	≥ 34	0.641	92.7	71.4	3.24	0.10	89.9	78.1	86.3
ZJU index	0.903 (0.843-0.962)	≥ 37	0.688	91.7	77.1	4.01	0.11	91.7	77.1	87.0
FSI	0.912 (0.849-0.954)	≥ 15	0.688	91.7	74.3	4.01	0.11	91.7	77.1	87.8
TyG index	0.902 (0.839-0.966)	≥ 8.5	0.649	90.6	74.3	3.52	0.13	90.6	74.3	86.3
VAI	0.857 (0.774-0.937)	≥ 1.49	0.675	87.5	80.0	4.38	0.16	92.3	70.0	85.5

¹The optimal cut-off with the highest Youden index.

AUC: Area under the receiver operating characteristic curve; 95%CI: 95% confidence interval; FLI: Fatty liver index; HSI: Hepatic steatosis index; FSI: Framingham steatosis index; TyG: Triglycerides and glucose; VAI: Visceral adiposity index; SE: Sensitivity; SP: Specificity; PLR: Positive likelihood ratio; NLR: Negative likelihood ratio; PPV: Positive predictive value; NPV: Negative predictive value.

Supplementary Table 1 presents the frequency and percentage of subjects categorized by: (1) Those below vs meeting the optimal cut-off for each steatosis biomarker to detect NAFLD; (2) Those with NAFLD as indicated by MRI-PDFF ≥ 5%; and (3) Those without NAFLD as indicated by MRI-PDFF < 5%. For the FLI, 33.6% (n = 44) of subjects fell below the cut-off for NAFLD, 34.1% (n = 15) of whom had NAFLD. Likewise, among the 23.7% (n = 31), 26.0% (n = 34), 25.2% (n = 33), 26.7% (n = 35), and 30.5% (n = 40) of patients who fell below the cut-offs of the HSI, ZJU index, FSI, TyG index, and VAI, respectively, 22.6% (n = 7), 23.5% (n = 8), 21.2% (n = 7), 25.7% (n = 9), and 30.0% (n = 12) had NAFLD, respectively.

Diagnostic accuracy of the previously established optimal cut-offs of biomarkers for the detection of NAFLD as defined by MRI-PDFF ≥ 5%

In addition, we analyzed the diagnostic accuracy of the previously established optimal cut-offs, i.e., ≥ 23 for FSI[23], ≥ 8.5 for TyG[24], and ≥ 1.25 for VAI[18], respectively (Supplementary Tables 2 and 3). In terms of optimal cut-offs, the value we determined in this study was lower than the established cut-off value for FSI (15 vs 23), while it was higher than the established value for VAI (1.49 vs 1.25). Notably, the optimal TyG index cut-off we determined in this study was the same as the established cut-off (≥ 8.5)[24]. The sensitivity of the established optimal cut-offs for VAI (92.7%) was higher than 87.5% (VAI ≥ 1.49) in this study, while the specificity of the established optimal cut-offs for VAI (71.4%) was lower than 80.0% in this study. However, the accuracy of the established and current cut-offs was similar for VAI (Table 3 and Supplementary Tables 2 and 3). The optimal cut-off for FSI identified in this study demonstrated higher sensitivity and accuracy (91.7% and 87.8%, respectively) compared to the previously established cut-off (78.1% and 80.2%, respectively), although it had lower specificity (74.3% vs 85.7%).

Diagnostic accuracy of the dual cut-offs of biomarkers to exclude or confirm the presence of NAFLD as defined by MRI-PDFF ≥ 5%

The methodology for selecting dual cut-offs resulted in similar sensitivity for the lower exclusionary cut-offs and specificity for the higher confirmatory cut-offs across the six noninvasive biomarkers in Table 4. Nevertheless, the percentage of subjects whose NAFLD was excluded, confirmed, or in between the dual cut-offs differed across these noninvasive biomarkers (Supplementary Table 4). Namely, far fewer subjects fell between the dual cut-offs for FLI (49.6%; n = 65), HSI (32.8%; n = 43), ZJU index (45.0%; n = 59), and FSI (46.6%; n = 61) vs for TyG index (59.5%; n = 78), and VAI (68.7%; n = 90). The percentage of subjects between the exclusionary and confirmatory cut-offs for NAFLD was similar across all six steatosis biomarkers: FLI (78.5%, n = 51), HSI (72.1%, n = 31), ZJU index (76.3%, n = 45), FSI (78.7%, n = 48), TyG index (75.6%, n = 59), and VAI (74.4%, n = 67; Supplementary Table 4).

Table 4 The diagnostic accuracy of noninvasive steatosis biomarkers in excluding or confirming the presence of nonalcoholic fatty liver disease as defined by magnetic resonance imaging proton density fat fraction ≥ 5%.

Biomarkers	AUC (95%CI)	SE, %	SP, %	PLR	NLR	PPV, %	NPV, %	Accuracy, %	Overall accuracy to exclude1 and confirm2 NAFLD, %
FLI (n = 131)
Exclusionary cut-off < 20	0.766 (0.684-0.836)	99.0	54.3	2.16	0.02	85.6	95.0	87.0
Confirmatory cut-off ≥ 68	0.701 (0.614-0.777)	45.8	94.3	8.02	0.57	95.7	38.8	58.8	48.1
HSI (n = 131)
Exclusionary cut-off < 33	0.781 (0.700-0.848)	99.0	57.1	2.31	0.02	86.4	95.2	87.8
Confirmatory cut-off ≥ 39	0.790 (0.711-0.857)	66.7	91.4	7.78	0.36	95.5	50.0	73.3	64.1
ZJU index (n = 131)
Exclusionary cut-off < 34	0.766 (0.684-0.836)	99.0	54.3	2.16	0.02	85.6	95.0	87.0
Confirmatory cut-off ≥ 41	0.732 (0.647-0.805)	52.1	94.3	9.11	0.51	96.2	41.8	63.3	52.7
FSI (n = 131)
Exclusionary cut-off < 7	0.775 (0.694-0.844)	97.9	57.1	2.28	0.04	86.2	90.9	87.0
Confirmatory cut-off ≥ 42	0.711 (0.625-0.787)	47.9	94.3	8.39	0.55	95.8	39.8	60.3	50.4
TyG index (n = 131)
Exclusionary cut-off < 8.0	0.700 (0.614-0.777)	100.0	40.0	1.67	0.00	82.1	100.0	84.0
Confirmatory cut-off ≥ 9.2	0.664 (0.576-0.744)	38.5	94.3	6.74	0.65	94.9	35.9	53.4	38.9
VAI (n = 131)
Exclusionary cut-off < 0.79	0.638 (0.549-0.720)	99.0	28.6	1.39	0.04	79.2	90.9	80.2
Confirmatory cut-off ≥ 3.29	0.617 (0.528-0.701)	29.2	94.3	5.10	0.75	93.3	32.7	46.6	29.0

¹Criteria for the exclusionary cut-off were sensitivity > 90% and negative likelihood ratio ≤ 0.1.

²Criteria for the confirmatory cut-off were specificity > 90% and positive likelihood ratio ≥ 10.

AUC: Area under the receiver operating characteristic curve; 95%CI: 95% confidence interval; FLI: Fatty liver index; HSI: Hepatic steatosis index; FSI: Framingham steatosis index; TyG: Triglycerides and glucose; VAI: Visceral adiposity index; SE: Sensitivity; SP: Specificity; PLR: Positive likelihood ratio; NLR: Negative likelihood ratio; PPV: Positive predictive value; NPV: Negative predictive value; NAFLD: Nonalcoholic fatty liver disease.

Diagnostic accuracy of the previously established dual cut-offs of biomarkers to exclude or confirm the presence of NAFLD as defined by MRI-PDFF ≥ 5%

In addition, we also analyzed the diagnostic accuracy of the previously established dual cut-offs for excluding and confirming NAFLD (i.e., < 30 ruled out and ≥ 60 ruled in for FLI[19], < 30 ruled out and ≥ 36 ruled in for HIS[20], and < 32 ruled out and ≥ 38 ruled in for ZJU index[22], respectively), as shown in Supplementary Tables 2 and 3. The study found that the exclusionary and confirmatory cut-off values for the HSI and ZJU index were marginally higher in previous research compared to the current study, whereas the exclusionary cut-off for FLI was lower and the confirmatory cut-off for FLI was higher than those identified in the current study. Although the sensitivities of the established exclusionary cut-offs for FLI (94.8%), HSI (100%), and ZJU index (100%) were similar to those in the current study (97.9%, 96.9%, and 99.0%, respectively), the specificities of the established confirmatory cut-offs for FLI (85.7%), HSI (71.4%), and ZJU index (80.0%) were significantly lower than those in the current study (94.3%, 91.4%, and 94.3%, respectively). Furthermore, the HSI and ZJU index yielded the best overall accuracy in excluding and confirming NAFLD, at 71.8% (i.e., a correct diagnosis in 94 of 131 subjects) and 67.9% (a correct diagnosis in 89 of 131 subjects) in the established study, respectively, and 64.1% (a correct diagnosis in 84 of 131 subjects) and 52.7% (a correct diagnosis in 69 of 131 subjects) in the present study, respectively (Table 4 and Supplementary Tables 2 and 3).

Diagnostic accuracy of biomarkers for estimating moderate-to-severe steatosis as defined by MRI-PDFF ≥ 11%

Figure 4B and Supplementary Table 5 illustrate the diagnostic accuracy of six noninvasive biomarkers in differentiating moderate-to-severe steatosis (MRI-PDFF ≥ 11%) from no/mild steatosis (MRI-PDFF < 11%). The diagnostic performance of all of these biomarkers in detecting moderate-to-severe steatosis was acceptable, though it was lower than their performance in detecting MRI-PDFF ≥ 5%. The HSI and ZJU index yielded the best AUCs of 0.857 (0.785-0.912) and 0.854 (0.781-0.909), respectively. The FLI, FSI, TyG index, and VAI yielded AUCs of 0.818 (0.741-0.880), 0.852 (0.779-0.908), 0.732 (0.648-0.806), and 0.704 (0.618-0.831), respectively. The sensitivity, specificity, PLR, NLR, PPV, NPV, and accuracy for each noninvasive biomarker at their optimal cut-offs for detecting moderate-to-severe steatosis are shown in Supplementary Table 5.

The AUCs for the FLI, HSI, ZJU index, and FSI were significantly greater than those for the TyG index and VAI (all P < 0.05). However, the AUCs for the FLI, HSI, ZJU index, and FSI did not significantly differ (all P > 0.05). Similarly, the AUCs for the TyG index and VAI did not significantly differ (P > 0.05).

Diagnostic accuracy of biomarkers for estimating severe steatosis as defined by MRI-PDFF ≥ 17%

The diagnostic performance of the six noninvasive steatosis biomarkers for estimating severe steatosis (MRI-PDFF ≥ 17%) vs no/mild-to-moderate steatosis (MRI-PDFF < 17%) is shown in Figure 4C and Supplementary Table 5. The diagnostic performance of these markers for detecting severe steatosis was also acceptable and similar to that for moderate-to-severe steatosis, but lower than that for detecting MRI-PDFF ≥ 5%. The HSI index and ZJU index yielded the best AUCs of 0.852 (0.779-0.908) and 0.851 (0.779-0.908), respectively. The FLI, FSI, TyG index, and VAI yielded AUCs of 0.829 (0.754-0.889), 0.832 (0.757-0.892), 0.730 (0.646-0.804), and 0.711 (0.626-0.787), respectively. The sensitivity, specificity, PLR, NLR, PPV, NPV, and accuracy for each noninvasive biomarker at their optimal cut-offs for detecting severe steatosis are shown in Supplementary Table 5.

The AUCs for the FLI, HSI, ZJU index, and FSI were significantly greater than those for the TyG index and VAI (all P < 0.05). However, the AUCs for the FLI, HSI, ZJU index, and FSI did not significantly differ (all P > 0.05). Similarly, no significant difference was observed between any two AUCs for the TyG index and VAI (P > 0.05).

DISCUSSION

This observational study performed a head-to-head comparison between six previously published noninvasive hepatic steatosis biomarkers and found that they exhibit strong diagnostic accuracy (approximately 0.90; range: 0.857-0.912) for detecting NAFLD as defined by MRI-PDFF ≥ 5%. Secondary analysis indicated that the AUC ranges for these biomarkers in detecting moderate-to-severe steatosis (MRI-PDFF ≥ 11%) and severe steatosis (MRI-PDFF ≥ 17%) were 0.704-0.857 and 0.711-0.852, respectively, suggesting their limited effectiveness in identifying moderate or severe steatosis. These data may enhance the clinical application of these models for the assessment of NAFLD and contribute to the development of a viable clinical method for the noninvasive detection of NAFLD. Furthermore, the use of these thresholds for diagnosing hepatic steatosis may change the design of clinical trials related to NAFLD therapy and decrease the costs of follow-up by diminishing the number of MRI scans for those trials using MRI-PDFF as the inclusion criterion.

While these hepatic steatosis biomarkers have been previously described, most of them were not validated by liver biopsy or MRI-PDFF. Some studies validated these biomarkers using only ultrasonography, which led to an underestimated NAFLD prevalence. FLI was originally validated in a cohort study using ultrasonography to diagnose fatty liver, achieving an AUC of 0.85 for detecting liver steatosis[19], comparable to our study's AUC of 0.899. Furthermore, Cuthbertson et al[32] validated that while FLI had an accuracy of 0.79 for hepatic steatosis as determined quantitatively by 1H-MRS, it could not predict hepatic fat content. In addition, it was also shown to predict the occurrence of new cases of NAFLD[33], which may facilitate early intervention. A Korean cross-sectional study[20] on ultrasound-detected NAFLD found that HSI is more accurate than conventional ultrasound for diagnosing fatty liver in patients with chronic hepatitis B, using liver biopsy as the reference standard[34]. However, external validation in other Asian populations is limited, and its correlation with MRI-PDFF has not been investigated. The ZJU index was developed and validated only in Chinese populations using ultrasound diagnostic criteria for NAFLD[22]. Recently, a study indicated that the ZJU index showed higher accuracy for diagnosing NAFLD compared to other established simple biomarkers based on western populations[35]. Moreover, the baseline ZJU index and its changes during follow-up independently predicted the risk of developing NAFLD in a Chinese population[36]. However, independent external validation is still lacking, and the role of the ZJU index in predicting hepatic fat content remains unclear. The FSI index was developed from a United States cross-sectional population in which 26.8% of patients were identified as having hepatic steatosis by computed tomography[23]. Later, the FSI index was well validated by two studies with Chinese populations, both of which showed good AUCs of more than 0.85 for predicting ultrasonography-diagnosed NAFLD[37,38]. However, the optimal cut-off for the FSI proposed by Long et al[23] was higher than both that used in the present study and the study performed by Shen et al[37].

The TyG index has recently gained attention as a reliable surrogate marker for identifying IR. Previous studies have shown that the TyG index is independently associated with hepatic steatosis in patients with chronic hepatitis C and NAFLD, aligning with our current findings[18,39]. A recent study by Zhang et al[24] demonstrated that the TyG index effectively identifies Chinese adults at risk for NAFLD, with an optimal detection threshold of 8.5. While the threshold aligns with our findings, the TyG index demonstrated significantly higher AUC, sensitivity, and specificity for diagnosing NAFLD in this study. The VAI index was derived from a Caucasian population and is a simple panel to measure visceral fat[40]. Few studies have explored whether VAI can independently predict NAFLD. Our study found a significant association between VAI and both the presence and severity of hepatic steatosis as measured by MRI-PDFF, aligning with findings by Fedchuk et al[18] and Vassilatou et al[41]. Furthermore, Petta et al[42] observed that steatosis biomarker levels were elevated in patients with advanced fibrosis compared to those with mild or no fibrosis and those with only steatosis. Conversely, Vongsuvanh et al[43] determined that VAI was not associated with fibrosis severity. Nonetheless, independent large-scale studies are required to further validate these data.

This study has several significant strengths. To our knowledge, this study is the first to demonstrate that six noninvasive biomarkers can effectively detect NAFLD in Chinese individuals as identified by MRI-PDFF, the most accurate noninvasive quantitative tool and a novel gold standard for assessing hepatic steatosis[6,13]. The rationale on MRI-PDFF as a surrogate gold standard of hepatic steatosis quantification is discussed in the Supplementary materials. Moreover, we synchronously analyzed the diagnostic accuracy of two types of cut-offs (single optimal cut-off and dual cut-off) for each biomarker and compared them with the accuracy of previously established thresholds for detecting hepatic steatosis, which contributed to a comprehensive understanding of these biomarkers.

Nonetheless, we recognize the following limitations. First, our study was limited by a relatively small sample size. As a non-interventional study, not all eligible subjects underwent MRI-PDFF, which may cause selection bias due to testing indications or patient preferences. Second, this study was conducted at a single tertiary hospital equipped with advanced MRI techniques that may not be accessible in other centers. Third, magnetic resonance elastography or liver biopsies were not performed. Thus, the definitive severity of the inflammation and fibrosis in the study subjects is unknown. However, the results of two simple and validated serum biomarkers (NFS and FIB-4) showed minimal or no fibrosis in the subjects. Hence, these fibrosis biomarkers may be limited in predicting hepatic steatosis with advanced fibrosis. Further studies will be conducted in the future to adjust for these confounders, such as fibrosis and inflammation based on histology. Additionally, the optimal cutoffs of this study lacked external validation, which may result in an overestimation of the diagnostic performance metrics. More external validation should be conducted in the future.

CONCLUSION

In conclusion, these simple noninvasive steatosis biomarkers were accurate for detecting NAFLD as defined by hepatic MRI-PDFF ≥ 5% in a Chinese adult population with no or mild fibrosis, which may be increasingly important in detecting and follow-up of NAFLD, even though they were unable to effectively detect moderate or severe steatosis. Additional studies are needed to validate the clinical application of these biomarkers in NAFLD diagnosis in multicenter, large-scale and long-term designs.